Home Assistant: must-have hardware, integrations worth setting up, and AI dashboards

Home Assistant is the brain. Everything else is an organ.#

Home automation is one of my favourite hobbies, and Home Assistant is the centre of it. Local-first where it can be, obsessively integrable, and flexible enough to tie lights, sensors, presence, media, energy, and odd IoT devices into one place that still works when a vendor cloud has a bad day.

The joy is not flipping switches from a phone. It is a house that feels thoughtful: fewer manual chores, better visibility, and a playground for people who like systems that respond to real life.

This post is the practical cut of that hobby. What hardware is actually worth buying for the house, which integrations repay the evening you spend on them, and how to use AI to build a dashboard view without fighting Lovelace YAML for three hours.

If you want the wider stack (Proxmox, media, energy, off-grid thinking), that lives in the broader homelab notes and on /homelab. For putting Home Assistant and the rest of the IoT mess on their own VLAN, see UniFi segmentation for Main, Guest, IoT and casting.

Start local, stay local#

My rule for the house:

• Prefer local protocols (Zigbee, Thread/Matter, ESPHome, wired Ethernet).
• Then local APIs on your LAN.
• Then cloud APIs as a last resort.
• For anything that matters (locks, heating, alarms, power), keep a manual fallback.

Cloud bridges are fine for a weekend experiment. They are a liability when the lights, heating, or door lock depend on someone else's outage calendar. Local integrations also update faster, which is what you want when an automation reacts to a door, a leak, or solar output.

Must-have hardware for the house#

You do not need a rack full of gear. You need a reliable brain, a radio that is not fighting Wi-Fi noise, and a small set of sensors that change how the house behaves.

1. The host#

Run Home Assistant on something that stays up.

• Home Assistant Green or a small dedicated mini PC if you want the simplest path and Home Assistant OS with apps (add-ons).
• A VM on Proxmox if you already have a lab. Home Assistant OS in a VM is the usual recommendation when you want apps, updates, and Thread or Z-Wave helpers without reinventing them.
• Docker works and is what I run in the lab inventory, but you trade away some of the OS-only app convenience. Pin the image version, put config on a volume, and treat upgrades like any other container.

Give it at least 2 vCPUs and 2 GB of RAM, put it on a UPS with the router, and back it up before every adventurous change. The host is Tier 1 kit: if it dies, the household notices.

2. A proper radio, on a cable#

Wi-Fi-only smart bulbs will get you started. A mesh radio is what makes a serious house.

• Home Assistant Connect ZBT-2 is the current Nabu Casa adapter for Zigbee or Thread, not both at once. Dedicate it to one role. Multiprotocol on one stick has been a poor experience, and they are clear about that.
• Use the USB extension cable. USB 3 ports and drives interfere with 2.4 GHz. Put the stick away from the NAS, the Wi-Fi AP, and the metal rack post.
• If you need Zigbee and Thread, buy two radios (or use a phone or Apple TV / Nest Hub as a Thread border router you already own) rather than hoping one dongle does both cleanly.
• Z-Wave still earns a seat for locks and some sensors in older UK installs. That wants its own stick and the Z-Wave JS path.

Buy a handful of mains-powered Zigbee or Thread routers early (smart plugs or always-on bulbs). Battery sensors need a mesh. A lonely coordinator in a cupboard will disappoint you.

3. Presence that is good enough#

Automations fail when the house does not know who is home.

• Phones via the Companion app (Wi-Fi, GPS zones, and notification channels).
• A couple of Bluetooth proxies or a room-level BLE setup if you want "which room" rather than just "home or away".
• mmWave or well-placed PIR for rooms where phones sleep in pockets and you still want lights to behave.

Start with home/away. Room-level presence is a second project, not a day-one requirement.

4. Sensors that pay for themselves#

These are the ones I would buy before another colour-changing bulb:

• Leak sensors under sinks, behind the washing machine, and near the boiler or stopcock.
• Door and window contacts on external doors and the garage.
• Temperature and humidity in the main rooms and the loft or plant cupboard.
• A smoke or CO path that you trust, even if that means leaving life-safety devices on their own certified system and only mirroring status into Home Assistant.

Fancy lighting is optional. Flood detection is not.

5. Cameras, if you want eyes#

If you care about security cameras, prefer wired PoE on the IoT or camera VLAN, with no internet for the cameras themselves. Pair them with local detection (Frigate below) instead of a cloud doorbell subscription you cannot automate properly.

6. A wall tablet is optional, not first#

A cheap Android tablet on the wall is brilliant once the automations exist. It is a terrible first purchase. Build useful automations and a phone dashboard first. The tablet is a display for a system that already works.

7. Energy hardware when you are ready#

A CT clamp, a Shelly EM-class meter, or a supported inverter / smart meter path turns Home Assistant into an energy brain. Do this after lights and presence, unless your whole reason for the project is solar and tariffs. Guessing at power with smart plugs alone gets old fast.

Integrations worth setting up#

Do not install fifty integrations on night one. Install the ones that unlock the next useful automation.

Core, almost everyone should have#

| Integration | Why it earns a slot | | --- | --- | | Mobile App | Presence, zones, actionable notifications, camera snapshots to your pocket. | | Person + zone | Turns device trackers into "Nathan is home" logic the rest of the house can use. | | Sun / weather | Cheap triggers for lighting, covers, and "is it actually dark outside". | | Backup | Built-in backups before upgrades. Non-negotiable. |

The local fabric#

| Integration / service | Why | | --- | --- | | MQTT (usually Mosquitto) | The bus for Zigbee2MQTT, Frigate, Tasmota, and a lot of DIY. Install it when something needs it, not because a blog said so. | | ZHA or Zigbee2MQTT | Your Zigbee mesh. ZHA is simpler inside HA OS. Zigbee2MQTT is the escape hatch when a device needs deeper options or you want the coordinator off the HA box. | | Matter and Thread | The path for newer certified gear without a brand bridge. Pair Thread credentials carefully; wrong preferred network is a classic foot-gun. | | ESPHome | When you want a sensor or switch that is yours, local, and flashable. Brilliant for one-off problems shelves do not sell. |

House and network#

| Integration | Why | | --- | --- | | UniFi Network (or your router equivalent) | Device trackers, WAN health, and "did that IoT thing appear on the wrong VLAN". Pairs well with the segmentation post. | | Ping / ping-based device trackers | Dumb but reliable for fixed machines: NAS, TV boxes, printers. | | Radio Browser / local media players | Small quality-of-life win on a kitchen dashboard. | | Cast / Sonos / Apple TV | Once discovery works across VLANs, media belongs on the same dashboard as lights. |

Energy and cars#

| Integration | Why | | --- | --- | | Energy dashboard | Home Assistant's built-in energy view. Feed it real kWh entities, not vibes. | | Powercalc (HACS) | Estimates usage for lights and devices that will never grow a meter. Useful filler, not a replacement for a grid sensor. | | Tesla / TeslaMate | If you drive one. Keep safety-critical charging logic from depending on a single flaky cloud API. Local where you can, cloud where you must, and never make "can I leave the house" depend on one vendor outage. |

Cameras and detection#

| Integration | Why | | --- | --- | | Frigate | Local NVR with object detection. "Person at the front door" beats "motion in zone 3". Needs MQTT. Prefer OpenVINO on an Intel iGPU for new modest installs unless you have a clear reason for another accelerator. | | Generic Camera / ONVIF | Fine for viewing. Weak for smart automations on their own. |

Frontend helpers (after the core works)#

| Piece | Why | | --- | --- | | HACS | Community cards and integrations. Install it when you have a specific gap, not as a lifestyle. | | Mushroom cards | Clean mobile-first cards that make a phone dashboard feel finished. | | Browser Mod or wall-tablet helpers | Only when you actually have a kiosk tablet. |

What I would skip early#

• Every brand bridge for bulbs you could have bought as Zigbee or Matter.
• Five overlapping presence systems before home/away is stable.
• Node-RED on day one if the built-in automation editor already covers the job.
• Cloud voice as the only way to control the house. Local Assist, or a physical switch, should still work.

Design the house in Home Assistant before you design the UI#

Dashboards fail when the data model is a mess. Before you ask any AI for cards:

1. Create floors and areas that match the real house.
2. Assign every device to an area.
3. Rename entities so a stranger could guess them: light.kitchen_ceiling, not light.hue_color_lamp_1.
4. Expose only what you need to voice assistants.
5. Decide which dashboards exist: Overview, Mobile, Energy, Security, maybe Lab.

AI is good at layout. It is bad at inventing a coherent naming scheme you will still like in six months. Do that part yourself.

How to use AI to create your dashboard view#

Lovelace is powerful and verbose. That combination is exactly where a language model helps: you bring the inventory, it drafts the YAML, you review and paste.

The safe workflow#

1. Take a backup (Settings → System → Backups).
2. Export or copy a list of entity IDs, friendly names, domains, and areas. Developer Tools → States is enough; some people dump a short Markdown table.
3. Prompt an AI with that list and a clear brief (examples below).
4. Ask for standard Home Assistant cards first so the result works without HACS.
5. Create a new dashboard (Settings → Dashboards), open Edit → ⋮ → Raw configuration editor, paste, save.
6. Fix broken entity IDs. There will be some.
7. Only then add Mushroom, sections polish, or theme tweaks.

Never point an agent with write access at a production Home Assistant and walk away. Review proposals. Prefer "generate YAML I paste" over "here are your admin credentials, go wild".

What to put in the prompt#

A good prompt is boring and specific:

• Goal: "Phone dashboard for daily use, not a TV wall."
• Layout: "One view per floor, or one home view with chips for rooms."
• Constraints: "Only these entity IDs. Use entities, button, gauge, thermostat, and picture-entity cards. No custom cards."
• Interaction: "Big lights and covers first. Sensors secondary. Hide unavailable entities if the card supports it."
• Platform: "Home Assistant Lovelace YAML for a sections or masonry view."

Then paste your entity table.

Example brief you can steal#

Build a Home Assistant Lovelace dashboard in YAML.

Audience: phone, one-handed, dark theme.
Views: Home, Downstairs, Upstairs, Energy, Security.

Rules:
- Use only the entity IDs listed below.
- Prefer built-in cards.
- Group by area.
- Home view: presence, outdoor weather, front door, alarm-style sensors, and scene buttons.
- Energy view: grid import, solar if present, and the top three high-load switches.
- Security view: cameras as picture-entity, door contacts, leak sensors.

Entity inventory:
(paste your list here)

In-Home Assistant AI options#

If you want the loop inside HA rather than ChatGPT in a browser:

• Community agents such as AI Agent HA or local panels like Kyber can discover entities and draft dashboards or automations for review. Treat them as power tools: useful, still beta-ish in spirit, and undeserving of unsupervised write access.
• MCP / API bridges let Cursor or Claude talk to a Home Assistant instance. Same rule: generate and review, or use a throwaway dashboard path first.
• Local models via Ollama are enough for YAML drafting if you do not want entity inventories leaving the house. Quality varies; your entity list quality matters more than the model brand.

Make the AI's job easier#

• Prefer areas over giant unsorted lists.
• Separate control entities from diagnostic entities in the inventory you paste.
• Ask for a mobile view and a wall view as two outputs, not one compromise.
• Iterate: "move climate to the top", "replace the entities card with a grid of buttons", "add a conditional card when nobody is home".
• When it invents fake entity IDs, push back with the inventory again. Models love confident hallucinations.

What "good" looks like#

A useful Home Assistant dashboard is quiet.

• One primary job per view.
• Lights and scenes reachable in two taps.
• Sensors visible without dominating the screen.
• Cameras on their own view so they do not thrash a phone on mobile data.
• The same area names you used in Settings, so voice, automations, and UI stay aligned.

If the first viewport needs a legend to explain itself, keep editing.

A sensible order of operations#

1. Host up, backups on, Companion app on your phone.
2. Radio on an extension cable, first router devices paired.
3. Leak and door sensors, then lights in the rooms you actually use.
4. MQTT only when Frigate or Zigbee2MQTT needs it.
5. Energy once you have a real meter path.
6. Areas and names cleaned.
7. AI-assisted dashboard draft, then a week of living with it on your phone.
8. Wall tablet last.

Summary#

Home Assistant rewards boring fundamentals: a stable host, a clean mesh radio, sensors that prevent damage, and integrations that feed presence, energy, and security into automations you notice every day.

The dashboard is the shop window, not the shop. Get the hardware and the entity model right, then let AI draft the Lovelace YAML while you stay in the approval seat. That split (you own the house model, the model owns the first draft of the view) is the fastest way I know to end up with a home screen that feels intentional instead of like a parts catalogue.

References#

• Home Assistant documentation
• Home Assistant Connect ZBT-2
• Home Assistant energy dashboard
• Frigate NVR
• Home Assistant Companion apps
• UniFi segmentation for Main, Guest, IoT and casting
• Containers versus VMs